It sort of reminds of mechanical engineering, where engineers sketch out the problem solution, do the calculations, and make sure the solution works. Then, they supervise drafters or technologists to produce the actual final product of engineering, drawings (2D/3D, paper/CAD, whatever).

Of course, that's in larger orgs. Smaller places will typically have an engineer doing everything.

Of course, there is overlap there. Technician->engineer used to be a fairly common career path.

You apparently have not encountered the "architecture astronaut" archetype[1]. Your final sentence stands; usually they can't.

[1] http://www.joelonsoftware.com/articles/fog0000000018.html

The issue with "architecture astronauts" is that most problems in software just aren't complex enough to warrant formal design. The core failing of these people is that they impose unnecessary design complexity in order to force the design to be complex enough to warrant formal design.

If the problem is inherently complex enough, particularly if it is multidisciplined, than it is the correct approach.

High-assurance sub-fields have many examples where the code is easiest part as it's just implementing a verified spec. Others have plenty of work on both sides. Altran (formerly Praxis) is a nice example with many commercial successes:

http://sdg.csail.mit.edu/6.894/dnjPapers/hall-correctness-by...

So called software architects seem to tend towards design only.

It's a "turtles all the way down" issue, though.

Software architects do the macro design, but they need to do that from an understanding of how it will be implemented, and the various lower level tradeoffs like consideration of the memory constraints of the target platforms.

In order to implement that design the non-architect programmers need to micro-design their bits that they build, too. And those smaller bits also often need abstractions/mini-architectures that aren't part of the macro architecture.

See proper engineering and the distinction between an engineer, a technician, and an assembler. There is overlap, of course, but designing and producing the product is a collaboration between all three. The engineer is primarily responsible for designing the system, the technician for helping the engineer translate the design into a realized system, and the assembler for actually realizing the system.

Well, but it's a bit weird with software. The "realized system" typically comes out of a compiler. When designing the system, the computer can often take your design and turn it into a realized thing.

In fact, we even have software called an assembler :)

Of course, it all depends on what level you're doing design at, and what level your language(s) work at. But the point is that those 3 roles aren't necessarily are fulfilled by humans when working with software.

I didn't say they necessarily were all fulfilled by humans. My point is that they are not necessarily all fulfilled by one human. Apropos of the comment I made in response to the "architecture astronauts", the complexity of most software projects is such that the design, from a technical standpoint, is a pretty small portion of the total effort required. In that case it makes sense for the technician to be the designer as well, which is something that often happens in the trades on simple jobs.

That does not mean that there can't or shouldn't be a distinction between the engineer and the technician on sufficiently complex projects. That is, they don't need to be filled by people who could perform either role, as the original comment I was replying to implied. That point was either miscommunicated in my original post or rubbed some people the wrong way.

The thing is that engineers produce and maintain documents. Specifically they produce source documents from which the end product has been derived.

That's exactly what a coder does.

There are some fields (draftsman, typist, others?) of people who created fine documents based on vague documents from engineers, architects etc. So in principle you could have a similar distinction between software archtects and coders. But it is exactly those document-gruntwork fields which have been transformed most by computers.

Are organizations still trying to move in that direction when engineering software products? I know it's been tried for decades now. But my limited experiences in organizations that tried it have not been very good. I'm curious if any organizations have made it work.

At the risk of drawing accusations of using the "No True Scotsman" fallacy, I'd say that most attempts have been trying to force-fit that type of style on projects that don't warrant it. That happens in all areas of engineering, though. The specific problem in software, IMO, is that too many projects are considered software "engineering" that shouldn't be, because they just aren't complex enough.